AI-assisted 3D analysis of grasping and reaching behavior of squirrel monkeys during recovery from cervical spinal cord injury.

We have previously demonstrated that machine learning-based video analysis, conducted via DeepLabCut, is more sensitive for detecting subtle deficits in hand grasping behavior than traditional end-point performance assessments. This superiority was observed in a nonhuman primate (NHP) model of cervical spinal cord injury, specifically a dorsal column lesion (DCL). The current study aims to further characterize the kinematic aspects of the deficits in hand reaching, grasping, and retrieving behavior from a 3D perspective following a DCL. Squirrel monkeys were trained to retrieve sugar pellets from eight wells, which were located either on a flat plate or a raised tube with varying well depths. This setup was designed to require coordinated finger movements during the task. Immediately after the DCL, the animals exhibited measurable behavioral deficits. These were characterized by significant increases in grasping speed squared and trial completion time, markedly widened movement trajectories of individual fingers, and abnormalities in inter-finger distance and orientation. Increased task difficulty was associated with more pronounced behavioral deficits. By three months post-DCL, video-based measurements indicated no significant recovery, even though global end-point performance had returned to baseline levels. Our findings demonstrate that deprivation of tactile information results in impaired dexterous hand behavior involving coordinated finger movements, and the impairment is sustained for 20 weeks. This spinal cord injury (SCI) model, along with DeepLapCut analysis, provides a valuable platform for separately evaluating sensory and motor functions and their contributions to dexterous hand behavior and may be used for evaluating therapeutic interventions using more sensitive behavioral outcome readouts.